A Conditional Value-at-Risk Based Portfolio Selection With Dynamic Tail Dependence Clustering

In this paper we propose a portfolio selection procedure specifically designed to protect investments during financial crisis periods. To this aim, we focus attention on the lower tails of the returns distributions and use a combination of statistical tools able to take into account the joint behavior of stocks in event of high losses. In detail, we propose to firstly cluster time series of stock returns on the basis of their lower tail dependence coefficients, estimated with copula functions, and secondly to use the obtained clustering solution to build an optimal minimum CVaR portfolio. In addition, the procedure is defined in a time-varying context, in order to model the possible contagion between stocks when volatility increases. This results in a dynamic portfolio selection procedure, which is shown to be able to outperform classical strategies.